Follow Slashdot blog updates by subscribing to our blog RSS feed


Forgot your password?
Check out the new SourceForge HTML5 internet speed test! No Flash necessary and runs on all devices. ×
Silicon Graphics

Tera Will Buy Cray Research 89

I just found this short news in C|Net which states that Tera Computer will buy Cray Research from SGI for an undisclosed amount of cash, stocks and notes (although the Wall Street Journal estimates the price that Tera pays is less then $100 million, which is a fraction compared to what SGI had to pay when they bought Cray - $740 million). Tera is going to change their company name to ... Cray.
This discussion has been archived. No new comments can be posted.

Tera Will Buy Cray Research

Comments Filter:
  • by EraseMe ( 7218 ) on Thursday March 02, 2000 @04:54AM (#1231663)
    I've been following SGI for quite a while now.. They began focusing on the x86 line as they figured the 64-bit Merced/Itanium/McKinney/[fill in the blank] would take over the marketplace sometime very soon for visual workstations.

    On the high end market, they purchased Cray, and continued work on the Orion line of supercomputers.

    Bad move buying Cray, bad move focusing on x86 64-bit imho. They had to get rid of Cray as it was becoming excess baggage. I may be wrong, but Intel is having trouble getting their 64-bit cpu higher than 400Mhz, while Alpha are laughing happily to the bank on their 21264 chipset, while working on their 21364 chipset (1000Mhz, 64-bit I believe). Sun continues working towards the UltraSparc-3 chipset (750Mhz 64-bit) and UltraSparc-5 chipset (1500Mhz 64-bit).

    Perhaps SGI should stick to high end 32-bit x86 chips for their workstation line, and continue pumping out MIPS cpu's on the beautiful Origin line. I understand fazeing out IRIX, but really..

    Where will it end? I love SGI products! Are they organized enough to provide me with a reliable long term gameplan, and be able to produce their servers in a timely manner?

  • Can anyone tell me why SGI is not traded on NASDAQ? Is it some company policy deal-io or what?
  • Maybe SGI don't see a future in old world big iron.

    These days anyone with the cash can buy a hundred or so fast workstations and using clustering build a *super* computer. I am inclined to believe that SGI has forseen the end of mainframes the amount of work they have put into linux filesystems (the really high end stuff) they must have encountered beowulfs and the like

  • by speed_bump ( 104415 ) on Thursday March 02, 2000 @05:15AM (#1231670)
    When it comes down to it, this really isn't too surprising given what's going on in the market and what happened to CRI as it grew into middle age. The market forces are obvious: lots of computing problems that used to be in the exclusive domain of supercomputers have been assumed by much smaller computers. Budgets of large government agencies have been dramatically reduced and with them, a large customer base has disappeared. If you're trying to make money, this spells disaster.

    Furthermore, like any organization, as they developed into more of a "sustained business" organization rather than a "let's build the fastest computer ever" organization, it became more and more difficult to innovate as rapidly as before (this is one of the reasons Seymour left). In the hey day of large government defense contracts, this was not a problem. However, as budgets dwindled they ran into significant difficulty penetrating new markets (a $30M machine is not an easy sell).

    It's really too bad as they have done some really neat things. In some senses it was the ultimate geek environment. In the engineering tradeoffs, speed always wins (which is why the things are so darn expensive). They used different wires for memory reads and writes, high memory interleaving, the I/O subsystems are multiple computers tasked for nothing more than I/O, vector registers make large FP computation chains very fast, no virtual memory (you can't use what you ain't got). In most cases, if it was slow they threw H/W at it. And then there was materials research for cooling and lots of other cool stuff.

    I don't know what Tera plans to do with them, but unless they have a good way to penetrate into the business markets (and there are opportunities), before long Cray will simply be a name for the history books. It's really too bad considering all the contributions they've made to the computing world.
  • I think this is good news. I've allways been fond of Cray, The company was Hq'd in my home town. The Cray name still has value even though the supercomputer market has dried up somewhat. I think the name change to Cray, reguardless of what processing market they're going after, will bring some level of branding with what the old Cray used to have. I'm glad Cray, in whatever form will still be around and not go the way of the dinosaurs
  • There was a big discussion of this under the "banning export of beowolf" heading a while back. The discussion, in a nutshell (hey, neat phrase... I should patent/trademark it!) pointed out that for problems which don't parallelize well, beowolfs won't do the job. One poster mentioned some problems from astrophysics, as examples which involve a LOT of interprocess communication, relative tot he number crunching. If you need to be able to handle large quantities of data, the PC architecture begins to fall down. If you need to sequentially do the same operations on every bit of that data, you need something fast and vectorized, with far better busses than PCI...

    Then there's the issue of reliability. In a discussion on big iron several days ago, it was pointed out that the big expensive equipment (mainframes, in that context) is made to far higher standards than the PC's. For the heavies, cost is no object. For the PC's, cost is the only object. There's only a one year warrenty, and if the manufacturer has to replace a few, who cares? It just means junior can't play his games for a few days.

    My point is that there are still things that we just can't do with PC's, not even with the nice four-way, crossbar-connected smp alpha from Microway that I've been drooling over. (It's only $13,000)

  • The details on the transaction do not seem to be public yet, the news stories I have read make a vague mention of "undisclosed amount of cash, stock and notes for the company."

    Given that TERA is one of those almost-on-life-support companies with some neato tech, perhaps this is really just a way for SGI to get access to TERA's technology. TERA certainly does not have the cash to buy Cray, so it seems likely that the majority of the transaction will be for shares and warrants of TERA. At the very least, such a transaction would give SGI a significant minority interest in TERA. On the other side of the equation, SGI gets to write-off a $600M+ tax-loss which should really help them out should they ever be able to sustain profitability.

  • As an irate IRIX user, let me tell you what is responsible for all of SGI's problems recently: a certain big fat penguin.

    While IRIX had advanced features that Linux users can only dream of, Linux has nothing to reccomend it to users. A few buggy desktop environments? A pretty window manager? Please. I can get those anywhere.

    Instead, it's the little things that I miss, such as support for mice with more than 3 buttons, or advanced virtual memory (which can often be stored on ram disks for better performence). When I was forced to switch to running Linux on my SGI, I soon found that nothing I had would run! How do they expect users to make the switch from a stable, open environment (in 3 years, IRIX has *never* crashed on me, but gcc dumps core all the time) to a buggy, hack-ridden kludge like Linux? Yeah, it's great that you let 16 year olds write your kernel drivers, but for me, I prefer leaving it up to the pros. Think you can call that kid for tech support? Not after 11:00, that's his bedtime.

    Ever since SGI turned from a high-power workstation producer to a sleazy Lintel vendor, I have been continually disgusted with them. Why would I want a dinky beige box running some toy OS? Hell, I once wrote a dynamic linker for my TRS-80, and I'm thinking I could extend it to full POSIX support. When I do, don't expect any open source from me. I prefer to actually eat occasionally, which I'm sure you'll understand once you leave your high schools or colleges and enter the real world.
  • SGI is findning that they are having a hard time competing in both software, hardware, and Super Computing markets all at the same time.

    This is just helping themselves get rid of one of those markets and focus on the ones that have the most groth.

    To be honest, with the ammount of support SGI is throwing into Linux, I wouldn't be surprised if IRIX disappears in the next few years and they wind up cutting Alias|Wavefront lose so they can just concentrate on building machines.

    Another good sign of this is the fact that SGI is working on a version of Maya (A|W's current top dog 3D modeler) for Linux.
  • ... And just the other day I was chuckling to myself on reading a William Gibson story which included a Cray computer...

    But then, it was a handheld; I doubt we're going to be seeing handheld Crays for a while...
  • Incidentally, it's a good idea to change the name "Tera", it links you to technology that will be dated when everybody focus on "peta-" things. It's like being named "20th Century Something" in the 21st century.
  • So, Cray was torn apart by SGI and Sun. Sun jerks over their portion and comes up with the Enterprise 10000. SGI flounders a while with their chunk and ends up only holding a name to sell.

    While I'm glad SGI is embracing Linux, in light of what they did to Cray, do we want the Linux name associated with this company? I would be happier if Sun truly supported Linux (instead of just paying lip service, community source... gack).

    In SGI's defense, $100 million is a pretty good price for a trademark and a domain name [].

  • I'd have to disagree with you. Linux is not the one that is thwacking SGI around. There marketing dept. is the only one to blame, (and maybe their old CEO who did that Visual PC fiasco...).

    Like you I love my Irix, it runs smooth and clean, never crashes on me and it's got all of the power functions that Linux has (and many more) but Linux hasn't got quite gotten down polish yet. But as a SGI supporter I feel embarassed to agree with you, your attitude leaves the same icky taste in my mouth that the Linux zealots leave (or Solaris, Mac, Windows....). If one were to look back in SGI's Irix history there are more than a few skeletons...
  • cray used to be a nifty company.. but are any of the people which were originally there still with them? i thought i heard somewhere that one reason the company failed miserably when the purchases started was because they canned all the good people.
  • Don't confuse mainframes with super-computers. Mainframes tend to be targeted towards the commercial market, and supers are sold to the technical market.

    Yes, there is a lot of overlap between the two markets nowadays, but they still have two different goals: commercial cares about reliabilty and uptime, technical cares about getting the right answers even if it takes a while (well, not too long).

    In the hey-day of supers, the joke was that super-computing was a synonym for unreliable computing. The reason the joke was true was two-fold: a small customer-base means less people to bang on the OS and find bugs, and because reliability was not a high-priority to the market.

    Sure, the hardware in supers tends to be over-engineered and rock solid, but in a cheapo-pc grade cluster situation you can easily make up for hardware reliability problems through redundancy, just have a couple or ten hot spares ready to pick up where the last node crashed.

    As for your point about problems with high inter-process communication requirements, I totally agree.
  • I have to agree with this fella, being an IRIX user of 2 years myself. It feels like a good time to use my farm analogy. My family owns and operates a large small-grains farm in the midwest. Our livelyhood depends on not just the weather, but our equipement as well. Most of our large equipment is John Deere and Flexicoil brand, between 2 and 6 years old, and under warranty. We don't use the latest and greatest wiz-bang toys, nor do we try to pull a 50-foot chisel plow with a Ford Ranger. When your business is on the line, or maybe even just your time, it only makes sense to use proven, relable equipement.
  • No.... SGI's been bleeding money for a while. They need to focus onto what they believe are their core businesses and get some profitability back. Then maybe their stock will rise and they can acquire other companies.

    So far as I know, a lot of the tech in the Origin series came directly from Cray. So... SGI paid a huge lump for access to some of cray's technology pool, but didn't really want to peruse Cray's

    Until we see MUCH better desktop and server systems, as well as gigabit ethernet, Beowolfs will really only be able to serve the lowend of the super computer market. Don't fool yourself into thinking that the technology is there yet.
  • Not all tech stocks are traded on NASDAQ. In fact, NASDAQ is a fairly "new" exchange (yeah, I know, almost 30 years old) and I believe it struggled at first. SGI was founded only (really, "only") 10 years after NASDAQ was and went public...I don't know when. At any rate, they probably chose to go with NYSE for some very good reasons at the time. And I don't know the ins and outs of switching exchanges or even if it's possible, but that could be a real SEC paperwork nightmare.

    A nightmare that SGI simply doesn't need.

    Is the exchange going to affect the stock? I don't know, but I kind of doubt it. SGI's stock sucks of its own accord, not due to being traded on NYSE. There's a chance they could be pulled up by the index of the exchange or something if they were traded on NASDAQ, but I doubt it would contribute enough to make a huge difference.
  • Yes, there are still several "old" (original) people there. I work in their processor group, and I've seen pictures some of the other employees standing with Seymour next to a Cray-2 :)

    "Cray, Inc.", huh? I guess they didn't have the $$$ to buy the "Research" part. :)

  • Well, SGI is committed to producing new MIPS machines through at least 2004 (? that may not be the exact number but I know it's about 4 years) at a minimum. If Merced flops, we will still have MIPS/Irix to sell :) In fact the machine under development releasing soon (SN1) will be able to use exactly the same hardware (I think you may need bigger power supplies with Intel) for everything except the processor cards which will be either MIPS or Intel. And it will scale to 512 processors on both! The above was not an official product announcement from SGI. Actual announcement may vary. Void where prohibited. Batteries not included.
  • They won't change to Linux. Not for at least 5 to 10 years. Because of the special architectures such as the SV1's MSP, they would essentially have to re-write the whole thing anyways.

    The UNICOS OS that Cray puts out is very highly tuned for performance and stability. The MTTI (Mean Time To Interrupt) on a J90 is close to a year of continuous uptime.

  • Do any of you know how I would go about obtaining the cabinet to a Cray X-MP? Not the computer, just the cabinet. You know, the one with a lovely built-in couch over the cooling unit. I've asked around, and I haven't gotten any good leads yet...
  • > may be wrong, but Intel is
    > having trouble getting their 64-bit
    > cpu higher than 400Mhz,

    AFAIK Intel plan to sell Itanium this summer with speed starting at 800 MHz.
    As for the performance, I don't know because the VLIW-like architecture needs a VERY clever compiler to generate a good level of ILP (Instruction Level Paralelism).

    We'll see, I'm sure there will be a huge number of benchmark, when it will be released.
  • My hometown is Chippewa Falls, WI, where Cray has
    all of their manufacturing facilities, so I've seen quite a few of the machines. I didn't think the X-MP was that comfortable, actually.

    Far cooler is the Cray 2 - the one that had the clear case and the waterfall heat exchanger for the florinert.

    As far as finding a cabinet, you might try comp.sys.super...
  • There was a J90 based desktop that fit into a double hight pizza box case. It was produced for the military. One CPU and a Florinert->heatsink->fan heat exchange unit. Very cool! (hot?)
  • CRI's CS6400 was not a storage system. It was an (up to) 64 CPU MPP system that used Sparc chips. CRI sold this technology to Sun when SGI aquired CRI. This was because the CS6400 directly competed with SGI's new Origin2000 line.

    Sun was the natural buyer. Sun turned it into their E10000 computer which was/is very powerful and often used as a database server or a transaction server. Check out the list of supercomputer sites and you will see it promenantly displayed.
  • SGI gutted Cray Research.

    They bought them to acquire some patents and interconnect technologies. On the high-end SGI Origin2000 the CPUs are connected by a high speed interconnect dubbed "CrayLink".

    They also got a hell of a good distributed storage management/migration system. This allows files that have not been accessed for a while to automatically be migrated from disk to tape. When the user opens the file, it is silently migrated back to disk without the user noticing (except for the latency). Works best with robotic tape systems such as STK.

    However, the apparent reason that SGI bought CRI was to eliminate the competition. Any customer requests for Cray systems were diverted by SGI sales over to Origin2000 systems. A customer had to try real hard to get a Cray. "No, I don't want the O2000, I really would prefer a T3E!"

    The T3E is a spectacular machine. It uses hundreds or thousands the very powerful Alpha chips with nearly linear scalability. The J90 & SV1 are a commercially successful line of systems that meet the needs of many customers. There is still life in these two architectures for sure.

    Tera is getting one of these lines. The T90/J90/SV1/SV2 line of vector computers.

    Tera is not trying to eliminate the competition as SGI did. Tera needs the name recognition and for their sake, they must also realize that Tera needs to adopt the pioneering culture that is still associated with everything Cray. So far, Tera has not done anything that makes us stand up and say "Wow!".

    Tera has interesting technology in the works, their "Multithreaded Architecture (MTA)". They have produced some very promising benchmarks (who hasn't). I just hope that they realize the great opportunity afforded them by purchasing the Cray name and get some genuine innovation going. They could also use a colorful spokesman, just make sure that he/she is second to the technology.
  • Amazing though there hardware is, SGI knows it simply can't compete with larger chip developers, especially Intel. Intel's R&D budget is orders of magnitude greater than anything the MIPS folks will ever see, and it's really starting to show. Right now, if you compare the fastest x86 chip with the fastest MIPS chip in integer performance, the x86 will absolutely blow it away. McKinley threatens to bring that same differential over to the FP side as well.
    So, yes, right now SGI workstations can hold their own with better FP and peripherals (bus, architecture, etc), but two years down the road it would be a very different story as IA64 kicks into gear. HP, whose PA-RISC has consistently been much faster than the MIPS line, realized this too in their shift towards IA-64. Let's just hope that SGI is successful in bringing all their technologies to Linux in that timeframe so they still have some workstations to sell.
  • by Animats ( 122034 ) on Thursday March 02, 2000 @10:02AM (#1231711) Homepage
    SGI bought Cray to get their big multiprocessors and clusters into the supercomputer market. They never really wanted Cray's operations at all. Cray had too many employees (I heard SGI managers complaining "What are we going to do with all those Cray people?") So it's no surprise they're selling the business.

    Besides, Cray-type supercomputers are over. All modern big machines are collections of single chip microprocessors. They have to be; the speed of light is too slow for a gigahertz big-box CPU. You can't even afford the slowdown for off-chip connections. Designing a fast CPU chip isn't worth doing unless you can sell a lot of them, so supercomputers have to use commodity CPUs. All that's left is architecting big multiprocessors and clusters out of commodity CPUs, which is mostly a software problem.

    SGI has other problems, the main one being lack of focus. "We're a graphics company". "We're a movie and game tool company." "No, now we're a server company". "We're an NT workstation company". "Forget that, now we're a Linux company". They keep trying to find some niche that isn't disappearing, and it's not working.

  • Usually the cases were reclaimed for scrap if you can believe it. I saw it happen to a just decommissioned X-MP in 1996. epaulson is right, the X-MP is not comfortable. The "couch" around it is steel and fiberglass, not padded.

    You are thinking of the Cray-1. It actually had its cooling pipes under a bench covered with padded naugahide. The story is that when someone suggested leather to Seymore, he replied to the effect "This computer isn't worth killing a cow over."
  • Hear, hear. An accurate assessment Maledictus.

    I got some SGI stock a couple years ago at a value price of $15. Two months later it went up to $30. I held it. Sigh.
  • Actually, it's not about hot spares. Systems in a cluster don't necessarily even have to have much of an identity.

    I've never used beowulf, but in the DQS system, you have a file that states the capabilities of the machines, like what class of box they are, how many processes you're willing to run on them, etc. Then there's an agent on each system which will run jobs. Big whoop, right? There's not much to that. Add in pmake (part of DQS, parallel make) and the BSD automounter, and you've got a cluster.

    Anyway, if a couple systems go down, no big deal, jobs just don't go to them. It's no stress. You just work on them, and put them back in, and when they come up, jobs start going to them again.
  • I'd like to add to the IRIX-bashing, since we had our extended linux bash, earlier.

    The only thing IRIX has that Linux doesn't have (yet? Or is it in the kernel now?) is a spiffy filesystem. I do recall SGI saying something about porting it to linux, though.

    On the other hand, until recently (or does it still?) IRIX had an xhost + in the Xconfig or what have you, so people could run xkeys and watch you type passwords. Go SGI! Then let's consider that the patches necessary to bring IRIX 5.x out of a security nightmare are something like 60mb in .tar.gz form. That's insane. Does that really point to a secure, stable OS?

    Linux is open source, but people still pay for it. The goal of linux is not to make money. It's to provide a fast, featureful operating system which runs on a variety of hardware. Porting software from one flavor of linux to another is a lot easier than porting from Solaris to IRIX (But then, IRIX has always been a problem child in that department, too.)

    Also, my redundancy approacheth, agreemsg on the whole intel killing SGI thing. Let's not forget, however, the 3d accelerator rush. When fast opengl-capable (Remember when OpenGL was basically a SGI buzzword?) video cards rolled out for PC platforms, that was what did it. If you're going to blame IRIX's death on something, blame it on the P54C chip, Windows NT 3.51, and the GLiNT graphics acceleration chips from 3dlabs (Remember, I'm talking about pricepoint here.) Now, you can get a GEforce 256 DDR card for $270 and it handles truly insane numbers of polys - So why use SGI even for that? Add to that the low cost of windows software, and BAM... you have no reason to go SGI. Sure they give you reliability (compared to a PC) but the price of the hardware is insane. Anyone recall how much a drive sled for the indigo R3000 ran you?
  • SGI bought CRI because at the time, SGI was doing very well and thought there would be synergies and CRI was in deep trouble because of (1) the reduction in the defense/security purchases (2) loss of sales to NEC/Fujitsu worldwide and extreme price pressures due to Japanese producer dumping of vector supercomputers and (3) difficulties in getting the T90 series to market as a side effect of (1) and (2). SGI got into trouble because they didn't keep far enough ahead of the market -- they didn't advance their technology as fast/faster than the commodity graphics products. So, rather than bailing CRI out, SGI took CRI down with them. There will always be a market for vector supercomputers. Vectors are far easier to program than non-vector MPPs SMPs, etc. and there are always problems that are harder than the most powerful computers and whose solutions have enough value that the $10-30 million price tag of a big vector system seems to be a bargain. Automobile, aerospace, pharmaceutical companies all understand the economics. Weather forecasting models have similar need for more computational power and economic value. When SGI got into trouble they should have turned CRI loose rather than drag it down. The announcement doesn't say what is going to Tera in the deal, e.g. Eagan or Chippawa Falls but so it is hard to speculate on whether enough of CRI will be intact to remake the national treasure that CRI once was.
  • "They also got a hell of a good distributed storage management/migration system. This allows files that have not been accessed for a while to automatically be migrated from disk to tape. When the user opens the file, it is silently migrated back to disk without the user noticing (except for the latency). Works best with robotic tape systems such as STK."

    IBM Mainframes have had this for years^H^H^H^H^Hdecades, though they do prompt the systems managers to swap tapes. Big deal.

  • I agree with your analysis of SGI, but you are completely wrong about the Cray computers. At least recent machines ARE collections of commodity CPUs. The Cray T3E is just a bunch of Alphas hooked together with an incredibly fast I/O system.

    It wasn't the individual processors that made Crays fast - I had a faster processor on my desktop than a single node in a T3E I have used. But hook up 700 of them with a bus that is faster than the one in your dektop system and watch out.

    Until networking components match the speed of these buses, most parallel computations will be much slower on beowulf type machines (commodity networking hardware) than on Crays.

  • A technical aside:

    How does swapping to a RAMdisk make any sense. I'm sure I'm missing something, but that just feels kooky. Unless you're meaning ramdisks on other computers (basically distributed shared memory) which assumes that net is significantly faster than local disk. I'll buy that for advanced networking. Just.

  • Who ever heard of Tera? But Cray....
  • It seems to me that everyone here is focusing on "The Demise of Cray". I've worked side by side with Cray employees and Tera ones. They're a perfect match, reguardless of who's funding what.

    Tera's revolutionary MTA has been proven to work very well. The best of both a vector and a MPP.
    Cray's SV1 may have been a flop, but the SV2 promises not to be. I guess I'm just interested in pure performance, and I, for one, can't wait to see what a combined Tera/Cray company pulls out.

    Imagine: Tera compilers for an SV2? or a Tera with all the I/O and experience of a Cray?

    excuse me, I think I'm drowning in my own drool...
  • Good comment, that (#18). Upmod it, pls.

    TERA does have some killer tech. I love that multithreaded switch on each clock stuff. Given that multi-threading is a big thing these days (cf sun's MAJC chip) and SGI has their CC-NUMA with non-uniform memory latencies, it makes a lot of sense for SGI to aquire TERA.

    TERA's chip would be the perfect complement for a CC-NUMA architecture. Perfect.

  • Yeah, I agree, big deal. Except when you consider that other Unix vendors do not have this feature and that IBM isn't licensing the technology.
  • Well, at least Gore Technology Group didn't get their hands on CRI as well as CCC. It would've been a shame to see all of that technology make it into a vault where no one could use it.

    It was a sad day in supercomputers when Seymour Cray died, and to help his legacy prosper, what did they do? They melted down all production CRAY-3's, and locked all the technology up tighter than fort knox. Hopefully Tera won't make the same mistakes as SGI.

    T h e U n d e r g r o u n d W a r e h o u s e

  • Hi Durinia, welcome to Tera^H^H^H^HCray Inc. ;-) What would you have done? I think the idea was to carve a distinct new identity and maybe imply that we would be doing development as well as research. We're pretty proud of our own research too ;-). Welcome to the team!
  • by John Carmack ( 101025 ) on Thursday March 02, 2000 @12:45PM (#1231728)
    I have been following both Cray and Tera for many years now. I have been saddened watching the last of the supercomputer companies wither and die. Supercomputers were always so COOL, but for most things, they just aren't so "super" any more.

    I have benchmarked several of my back end utilities on cray systems, and one of them on the early tera machine (the early compiler exploded on the others). None of the single processor runs were as fast as a pentium III, and this was quite some time ago.

    Understand that this was often branchy and recursive code running with only 3D sized vectors, so it isn't the sweet spot for traditional supercomputers. If I was doing nothing but multiplying 1k by 1k matricies of doubles, even a five year old cray would kick the crap out of the latest athalon. Unfortunately, none of my code looks like that.

    I even spent some time thinking how I could restructure calculations into a vectorizable form, which might make a cray J90 competative. I wanted to buy a Cray! Of course, this was silly. It took less effort to make the code SMP friendly, and the payoff was much larger.

    We wound up with a 16 processor SGI origin-2000 system, which has been easy to develop for and predictable in performance. We just recently bought an 8 processor Xeon, which is actually faster than our old 16 processor SGI, but at exactly one tenth the price (the downside is that it is maxed out, while the SGI still has tons of growth potential).

    I program all heavy workloads in a parallel fashion now as a matter of habit, but it is easy to overstate the benefits of parallel systems.

    The common lunux advocate position of "beowulf makes supercomputers obsolete" isn't quite right. Even with code that is already written in a parallel manner, there is a large difference between developing for a shared memory system and a distributed cluster. Developing a compute (and especially data) intensive program for a cluster rather sucks.

    If there was a single processor system that was really four times as fast as "consumer" machines, even if it cost fifty times as much, we would buy it. Unfortunately, there isn't. When the product release cycles are favorable, Alpha systems may be twice as good as x86 systems, but not much beyond that unless you are doing the 1k by 1k matrix type stuff.

    It is often forgotten that the original Cray-1 was largely a success because it was the fastest SCALAR processor of the time. The vectorization was just a bonus. Now, vectorization is the only thing that gives them a reason for existance.

    The tera architecture is very interesting, but for scalar code, it is VERY, VERY slow. I'm not sure if it will be competative with large processor count SGI origin-2000 systems even after it matures. It gains ease of programming from the lack of caches, but it gives away a lot of problem domains where it is going to look stupidly slow.

    If a supercomputer company could make a scalar processor that ran many times faster than existing processors and had similar SMP capabilities, it would probably be a success. Even if it cost a million dollars, filled a room, and burned hundreds of kwatts.

    The problem isn't really that supercomputers are bad, its just that we are so spoiled by how AMAZINGLY GOOD our cheap consumer hardware is.

    I do still worry about the stiffling of innovation that comes from having so few architectural directions for systems, but in the end, wall clock performance is what really matters.

    John Carmack

  • The only thing IRIX has that Linux doesn't have (yet? Or is it in the kernel now?) is a spiffy filesystem.

    SGI's X server kicks XFree's ass up and down the street. Not to disparage XFree: this is partly because the SGI X server itself is very, very good, but largely because SGI's graphics hardware is better than anything that you can buy in the PC world for love or money. I'm not talking about raw polygons/second in a full-screen window, I'm talking about use as a graphics-intensive desktop. There's nothing that compares to an O2, or even an Indy, as far as running an X desktop goes.

  • > you qualify for deeply discounted SGI software
    > and the Varisty Program

    You only qualify when you agree to take out
    a joint hardware/software support contract.
    This is the mighty sting in the tail - for
    what they wanted off me I could have upgraded
    to better Alpha hardware and DEC Campus.
  • Why bother with a supercomputer when you can do stuff like what they did with SETI@Home???

  • It's an old idea. IBM and others used to make special high speed disk drives for VM paging. When RAM got cheap enough, the mechanical drive assembly was replaced with RAM with battery backup. The interface was the same but the performance was much better. Adding the same amount of RAM to main memory may not be possible due to architectural limits or lack of expansion space in the CPU cabinet. Some systems support "bulk memory", which isn't directly addressable, but can be used for paging or caching files.
  • ...and also considering the fact that the Cray/SGI/Whatever version is the most robust and efficient version. Provided, you REALLY end up paying for it...
  • by Dogun ( 7502 )
    So the big red wavy box decides to buy off SGI? WHAT THE HELL!!! How messed up can the world get? Last time I heard of these guys, they stood for "Truly Enigmatic Rusty Artifact"
  • If you had the right combination of $$, you could see if the Chippewa Falls technology museum would sell theirs. =o)

    Oh, and speaking as a Dec `99 Madtown graduate.. Go Badgers!
  • Just a comment about what you write on "T3E scales almost linearly". That's something in favour of Cray machine, but you're forgetting that obtaining this linear scalability requires quite hard programming (remember that the T3E is a distributed-memory machine and, although you could use the shmem routines to do some kind of shared memory programming, this is quite a high-level solution). On the other hand, you've got this CC-NUMA SGI architecture, which allows you to program the whole huge machine as a SMP. In brief, the T3E is a native message-passing computer, while the Origin is actually a very sophisticated (far more than Sun's ones) SMP. This is definitively a point in favour of Origin product-line.
    But, OK, the Cray is also a great machine.
  • I've used SGI's for a number of years, including recently Power Challenges, Octanes, and such like.

    Yes, the graphics are good. But the processing stinks. On most of the codes I run (mix of 32-bit fp and integer array index, FFTs and such like), a Pentium II/300 will outperform an R10000 with heaps of cache and memory bandwidth.

    The ccNUMA + crossbar stuff is great, but needs a faster processor to go with it.

    Still, they are doing great work with Linux. For their software work I hope they do well.
  • This is probably too late to be read by_______ anybody, but oh well._____________________________ __________________________________________________

    According to , SGI isn't selling all of Cray.__ Instead they're just selling the vector lines,____ and keeping the parallel lines (BTW try not_______ to cringe too much at the scaler/scaler mix-up____ in the article).__________________________________ __________________________________________________

    So SGI is keeping the T3E__ line, which has most of Top 500 Supercomputers Computers [] that Cray still had on the list. Plus, the T3E line meshes much more closely with the SGI Origin line anyway.

    So SGI's sale of Cray for $100 Million is not quite as bad as it seems.

  • Sorry about the other garbled version.

    According to this article [], SGI isn't selling all of Cray Instead they're just selling the vector lines, and keeping the parallel lines (BTW try not to cringe too much at the scaler/scaler mix-up in the article).

    So SGI is keeping the T3E line, which has most of Top 500 Computers [] that Cray still had on the list. Plus, the T3E line meshes much more closely with the SGI Origin line anyway.

    So SGI's sale of Cray for $100 Million is not quite as bad as it seems.

  • Wrong. Don't believe everything you read in trade journals. The deal does include the T3E line as well as the installed base. There may be restrictions on further development, though, I'm not privy to such info. See for more info.
  • I have to agree, at least, not in their current form.

    Granted, their journaling file system is nice, and all, and their contributions are nice - a company still has to make money, and I don't see how their buisness model can be viable.

    Just my $0.02 CDN though.

  • Thanks for the post, AC. It's great to see perspectives from both sides of the line.
  • I used to work for a company in the UK called Cray Communications (part of Cray Electronics). I think it was bigger than Cray Research, but obviously people got confused... ("You work for Cray?")
    I left them ages ago. They are now called Anite.
    (Sorry for the slightly off topic post - I thought you *might* be interested)

Anything free is worth what you pay for it.